Being able to accurately predict future levels of port congestion is of great value to both port and ship operators. However, such a prediction tool is currently not available. In this thesis, a Long Short-Term Memory Recurrent Neural Network is built to fulfill this need. The prediction model uses information mined from Automatic Identification Systems (AIS) data, vessel characteristics, weather data, and commodity price data as input variables to predict the future level of congestion in the port of Paranaguå, Brazil. All data used in this study are publicly available. The predictions of the proposed model are shown to be promising with a satisfactory level of accuracy. The conclusion and evaluation of the presented model are that it serves its purpose and fulfills its objective within the constraints set by the authors and its inherent limitations.

nhhmas